lewinfox / lambdar Goto Github PK

Because we can specify the handler when running the container (docker run <image> filename.function_name) there's no reason not to allow multiple instances of the same container, each running a different function. Saves building multiple different images.

As given in this example we can add CMD ["file.function"] into the Dockerfile rather than adding it to the docker run command.

When generating _lambdar.yml, look for @lambda tags and populate the YAML with the right file.function handler.

There should only be one @lamba tag per project - check and handle this.

In the lambda_runtime.R we can use

  response_object <- withCallingHandlers(
    {
      res <- list(
        result = NULL,
        status = "ok"
      )

      lambda_function <- match.fun(function_name)
      res$result <- callr::r(lambda_function, args = event_content)
      res
    },
  # ...

to run the lambda in a separate subprocess. We still get errors back but we lose visibility of warnings. Is this desirable? Given that we're not doing much with warnings is this an acceptable tradeoff for extra safety?

Look for library() and pkg::fun(). Compare against the list of packages known to be installed in the container (how?).